AI integration
AI integration checklist: data, permissions and human review
Before adding an AI feature to a website or business tool, define what it may read, suggest and change. This checklist helps you prepare a project brief and identify decisions that need an owner before implementation begins.
Describe one user and one task
State the situation in ordinary language: who needs help, what question they have and where the answer will appear. Specify what is outside the feature's scope. An internal drafting tool and a customer-facing assistant need different review and access decisions. Confirm whether your existing software exposes the interfaces needed for the proposed connection before committing to a particular design.
Checklist
- Name the user and the desired outcome.
- List systems and account owners.
- Separate required capabilities from optional ideas.
Approve the information sources
Inventory the documents or records the feature needs. Decide which version is authoritative and who updates it when the business changes. Check the proposed provider's handling and retention settings before sharing private information. Use a small, approved sample to investigate feasibility. Do not treat access to an entire shared drive as a substitute for choosing relevant sources.
Checklist
- List approved sources and update responsibilities.
- Remove unneeded personal or confidential information.
- Review access and provider settings before connection.
Separate answers from actions
Decide whether the feature only retrieves information, drafts a response or can change something. Give each connection only the permissions necessary for the agreed purpose. Ask what should happen if a user requests information they cannot access. Customer-facing answers should have a path to a person when the approved material does not support a useful response.
Checklist
- Define read and write permissions separately.
- Require approval for consequential actions.
- Document escalation and a manual fallback.
Agree on tests and ownership
Prepare expected behavior for typical questions, missing facts and out-of-scope requests. Test source updates and integration failures, not just successful examples. Identify who reviews problems after launch and how the feature can be disabled. Confirm subscriptions, usage charges and ongoing responsibilities so the integration can be supported after the initial implementation.
Checklist
- Use realistic normal and failure cases.
- Set acceptance checks for accuracy and response time.
- Assign maintenance, cost and incident ownership.
A decision log for your project brief
Illustrative planning worksheet. The entries are examples, not requirements for every integration.
| Decision | Example to discuss |
|---|---|
| Source of truth | An approved service document maintained by the business owner. |
| Allowed behavior | Draft an answer with a link to the relevant source. |
| Uncertain answer | Ask a clarifying question or route the request to a person. |
| Review owner | A named staff member checks test cases and source updates. |
Turn your notes into a project brief
Share your goal, existing tools or website, relevant examples and the questions you still need to resolve. Do not email passwords or private customer records.
Related guides
AI automation
AI workflow automation for small businesses: where to start
Choose a practical first AI workflow for your small business. Map the task, compare rules with AI, define human review and plan a limited pilot before scaling.
Read the guideAI evaluation
How to evaluate an AI automation pilot
Evaluate an AI automation pilot using baseline task time, review effort, corrections, quality checks and operating costs, with an illustrative planning calculator.
Read the guide